Hunyuan-DiT: A Powerful Multi-Resolution Diffusion Transformer with Fine-Grained Chinese Understanding

TL;DR

Hunyuan-DiT is a multi-resolution diffusion transformer that achieves state-of-the-art Chinese-to-image generation and supports multi-turn multimodal dialogue, combining fine-grained language understanding with advanced image synthesis.

Contribution

The paper introduces Hunyuan-DiT, a novel diffusion transformer with a new data pipeline and multimodal language model for improved Chinese image generation and dialogue capabilities.

Findings

Sets new state-of-the-art in Chinese-to-image generation

Demonstrates effective multi-turn multimodal dialogue

Achieves high-quality image refinement based on context

Abstract

We present Hunyuan-DiT, a text-to-image diffusion transformer with fine-grained understanding of both English and Chinese. To construct Hunyuan-DiT, we carefully design the transformer structure, text encoder, and positional encoding. We also build from scratch a whole data pipeline to update and evaluate data for iterative model optimization. For fine-grained language understanding, we train a Multimodal Large Language Model to refine the captions of the images. Finally, Hunyuan-DiT can perform multi-turn multimodal dialogue with users, generating and refining images according to the context. Through our holistic human evaluation protocol with more than 50 professional human evaluators, Hunyuan-DiT sets a new state-of-the-art in Chinese-to-image generation compared with other open-source models. Code and pretrained models are publicly available at github.com/Tencent/HunyuanDiT